Not 100% sure if this is the right place for this question, so redirect me if needed.
I was listening to a podcast about disinformation due to AI (80 000 Hours podcast, #96 with Nina Schick), which focuses on so-called “deep fakes” where an AI copies a person’s likeness and voice and creates a scenario with them that never happened. This could have big effects on the ability to use photography and video as evidence. One suggested method of combating this is to make a workflow that authenticates the product, i.e. encrypted metadata is attached automatically when a picture or video is taken, and as soon as the image is manipulated then parts of the metadata change, indicating that something could be wrong with it. Adobe has made one such system, called the “Content Authenticity Initiative”.
As soon as I heard this, I thought; “couldn’t someone who wanted to use this system to pass off a faked video as real, first make a deep fake the normal way, display it on a screen or project it on something, and then film that using the authenticity system? They would then have a faked video with all the metadata saying that the video has not been manipulated.”
I know that it is easy to seamlessly re-record audio, but the thing I have insufficient knowledge about, and would like some expert opinion on, is if it is practically possible to do this for video. I.e. can you have a screen or projection showing a video, and film it in such a way that you cannot tell that it is a screen/projection, as opposed to being direct filming of the subject/scene? I tried to search for potential issues but could not find anything that a knowledgeable photographer couldn’t easily fix.
Thanks in advance!
I was listening to a podcast about disinformation due to AI (80 000 Hours podcast, #96 with Nina Schick), which focuses on so-called “deep fakes” where an AI copies a person’s likeness and voice and creates a scenario with them that never happened. This could have big effects on the ability to use photography and video as evidence. One suggested method of combating this is to make a workflow that authenticates the product, i.e. encrypted metadata is attached automatically when a picture or video is taken, and as soon as the image is manipulated then parts of the metadata change, indicating that something could be wrong with it. Adobe has made one such system, called the “Content Authenticity Initiative”.
As soon as I heard this, I thought; “couldn’t someone who wanted to use this system to pass off a faked video as real, first make a deep fake the normal way, display it on a screen or project it on something, and then film that using the authenticity system? They would then have a faked video with all the metadata saying that the video has not been manipulated.”
I know that it is easy to seamlessly re-record audio, but the thing I have insufficient knowledge about, and would like some expert opinion on, is if it is practically possible to do this for video. I.e. can you have a screen or projection showing a video, and film it in such a way that you cannot tell that it is a screen/projection, as opposed to being direct filming of the subject/scene? I tried to search for potential issues but could not find anything that a knowledgeable photographer couldn’t easily fix.
Thanks in advance!
